cumsum! does not check array sizes

[stevengj]

julia> cumsum!([0,0], 1:4)

signal (11): Segmentation fault: 11

[stevengj]

Also, it looks like the code assumes 1-based indexing; PR forthcoming that should fix both this and #18336.

[stevengj]

Another problem:

julia> cumsum(Any[])
ERROR: BoundsError: attempt to access 0-element Array{Any,1} at index [1]
 in cumsum(::Array{Any,1}) at ./arraymath.jl:448

which comes because the _cumsum_type(v) = typeof(v[1]+v[1]) definition is wrong. Grr.

[stevengj]

And cumsum(Any[1, 2.3]) throws an InexactError. I think _cumsum_type is just wrong and we should use similar(v) in all cases.

[andreasnoack]

Bools are not stable under addition.

[stevengj]

Oh, right.

[nalimilan]

promote_type(+, ...)?

[stevengj]

We already have an r_promote that is used for reductions; it seems better to be be consistent with that. In any case, please move this discussion to #18364.